In this paper, a perception net based method is proposed to estimate upcoming lane and vehicle pose for automatic lane keeping. The road is modeled by three connected rectangular plates so as to give computational efficiency and robustness in real time sequence. Uncertainties in estimating plate pose parameters and measuring the lane are considered and propagated forward and backward through the perception net to reduce the uncertainties of the resultant plate pose parameters. The constraint condition that the lane has constant width and is always parallel to the plate is imposed to the optimization problem in the perception net module to adjust plate pose parameters. As a result, vehicle pose and lane to follow under large noise as well as variants embedded in the street images will be estimated. An experimental result is presented, where a series of pictures of the real lane as well as synchronized information on vehicle movement are taken and the proposed method is applied to the real situation. For the covering abstract see IRRD E102946.


  • English

Subject/Index Terms

Filing Info

  • Accession Number: 00781130
  • Record Type: Publication
  • Source Agency: Transport Research Laboratory
  • ISBN: 89-950073-2-X
  • Files: ITRD
  • Created Date: Jan 7 2000 12:00AM